Research Article Age-Related Evolution Patterns in Online Handwriting Gabriel Marzinotto, 1 José C. Rosales, 1 Mounîm A. EL-Yacoubi, 1 Sonia Garcia-Salicetti, 1 Christian Kahindo, 1 Hélène Kerhervé, 2,3 Victoria Cristancho-Lacroix, 2,3 and Anne-Sophie Rigaud 2,3 1 SAMOVAR, Telecom SudParis, CNRS, University of Paris-Saclay, Palaiseau, France 2 AP-HP, Groupe Hospitalier Cochin Paris Centre, Hˆ opital Broca, Pˆ ole G´ erontologie, Paris, France 3 Universit´ e Paris Descartes, EA 4468, Paris, France Correspondence should be addressed to Mounˆ ım A. EL-Yacoubi; mounim.el yacoubi@telecom-sudparis.eu Received 19 February 2016; Accepted 14 August 2016 Academic Editor: Pietro Cipresso Copyright © 2016 Gabriel Marzinotto et al. Tis is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Characterizing age from handwriting (HW) has important applications, as it is key to distinguishing normal HW evolution with age from abnormal HW change, potentially triggered by neurodegenerative decline. We propose, in this work, an original approach for online HW style characterization based on a two-level clustering scheme. Te frst level generates writer-independent word clusters from raw spatial-dynamic HW information. At the second level, each writer’s words are converted into a Bag of Prototype Words that is augmented by an interword stability measure. Tis two-level HW style representation is input to an unsupervised learning technique, aiming at uncovering HW style categories and their correlation with age. To assess the efectiveness of our approach, we propose information theoretic measures to quantify the gain on age information from each clustering layer. We have carried out extensive experiments on a large public online HW database, augmented by HW samples acquired at Broca Hospital in Paris from people mostly between 60 and 85 years old. Unlike previous works claiming that there is only one pattern of HW change with age, our study reveals three major aging HW styles, one specifc to aged people and the two others shared by other age groups. 1. Introduction Handwriting (HW) analysis has recently been investigated for detecting pathologies and cognitive decline [1–3]. In this context, age characterization from HW [4–6] is fundamental as it may allow distinguishing normal HW change due to age from abnormal one, potentially related to a cognitive decline. In this paper, we address the problem of age characterization from online HW. Te goal is to detect HW styles and study their correlation with age, by the analysis of spatiotemporal HW parameters. Several previous studies have tackled the problem of age characterization of healthy persons from both ofine and online HW. Sometimes, this characterization is carried out by visual inspection [2, 3, 7–9] through observable features as, for example, letter size and width, slant, spacing, legibility or smoothness of execution, alignment of words with respect to baseline, and number of pen lifs. On the other hand, sometimes it is carried out by extracting automatically features from the ofine raw signal [10] or from the raw temporal functions of online handwriting using a digitizer [4–6, 11, 12]. HW style characterization has been widely studied for both online [13] and ofine [14] recognition tasks, and it is used to design writer style-dependent recognition models. Inference of HW styles, however, is difcult as there are no rules to defne a HW style. A clustering algorithm is thus usually required (Gaussian Mixture Models [14], -means [15], Self-Organizing Maps [13], Agglomerative Hierarchical Clustering [16], etc.). Previous works for clustering HW styles tackled the problem at the stroke level [16], character level [15], or word level [17]. We believe, however, that style characterization should rely not only on this raw signal information but also on high-level information associated with the variability observed across writer words. Previous works on the correlation between age and HW agree that age leads to a diferent behavior of the features Hindawi Publishing Corporation Computational and Mathematical Methods in Medicine Volume 2016, Article ID 3246595, 15 pages http://dx.doi.org/10.1155/2016/3246595